Can AI agents automate Microsoft Defender XDR?

Microsoft Defender XDR is automatable today through Microsoft Graph security APIs and native Defender XDR REST. This record is the unified XDR suite formerly documented as Microsoft 365 Defender. It is not Microsoft Sentinel, CrowdStrike Falcon, IBM QRadar, or Wiz. Graph lists and updates incidents at GET/PATCH https://graph.microsoft.com/v1.0/security/incidents with delegated or application SecurityIncident.Read.All or ReadWrite.All. Native incidents and hunting use https://api.security.microsoft.com/api/ and OData along the /api path. Microsoft tells callers to prefer Graph; the older hunting endpoints at api.security.microsoft.com stop returning data on 1 February 2027. Authentication is OAuth 2.0 against Microsoft Entra ID. User-context apps use an authorization-code flow. Application-context daemons use client credentials against resource https://api.security.microsoft.com, with Microsoft Threat Protection application permissions such as Incident.Read.All. Advanced hunting runs KQL over up to 30 days of raw signals, with per-tenant CPU and 45-calls-per-minute floors. Coverage is partial because the documented surfaces are incidents, hunting, and streaming, not the entire Defender portal. A first-party Defender XDR MCP server is not established on the opened pages. Sentinel's hosted MCP can expose Defender triage tools after lake onboarding; that server belongs to the Sentinel identity. Connector catalogs were not opened. Computer-use viability is unassessed. API scores 8 and is the best path.

Best path today: api · Overall automatability: 8/10